Research On Mechanical Behavior Of Reinforced Concrete Special-Shaped Column Joints Based On Machine Learning

In order to research RC special-shaped column joints by data-driven instead of mechanism-driven,This research provides effective theoretical and technical support for the seismic resistance of RC special-shaped column joints,so that the value of RC special-shaped column frame structure can be fully utilized in civil engineering.The research is as follows:(1)Based on domestic and foreign literature data,experimental data and finite element data,the data set of the joints of RC shaped columns is established and its related input and output variables are determined.The data set includes 13 parameters,including beam and column section size,column height,beam length,reinforcement diameter,concrete material strength,steel reinforcement material strength,stirrup ratio,axial compression ratio,ultimate bearing capacity,etc.and perform feature analysis and data normalization on the heterogeneous column joints data set.According to NFL theory,The BSA algorithm is superimposed on the neural network algorithm to establish a prediction model of RC special-shaped column joints,The number of implied layers,the amount of the neuron,the activation function and another parameters are identified,and the training and bearing capacity prediction results show that the improved BSA-BP neural network algorithm has high accuracy in prediction.(2)Based on the low cycle loading experiment of RC special-shaped column side joint,the process of crack development and the final failure form were analyzed.During the experiment,high-definition resolution photos were collected with a digital camera,the photos were cut and manually labeled as cracks and no cracks to establish a data set.VGG16 and Mobile Net convolutional neural network algorithm model were used to establish the computer vision model,it has a high accuracy after training.This method eliminates the need for human extraction of pixel points in the image,thus effectively reducing human error.(3)Based on SHAP interpretable machine learning theory,In this passage,the xgboost algorithm is used to establish a data set of RC special-shaped column joints as a research object,and the xgboost algorithm is used to perform a non-human subjective analysis of the factors affecting the RC special-shaped column joints,and to interpret the load capacity prediction of the same and different size RC special-shaped column joints.The training model has its own explanatory ability to analyze the quantitative Shapley of single-sample features at two levels,macro and local,to obtain the factors affecting the joints load-bearing capacity.